Optimizing Disease Identification in Mango Leaves: A Channel Attention Approach with Enhanced EfficientNet

In the GDP of Bangladesh, the contribution of agriculture is significant. To the mass population, mango is the most preferred fruit because of its variety, nutritional value, and taste. However, mango production is hampered due to leaf disease often. So, automatic mango leaf disease diagnosis methods are essential, especially in the context of Bangladesh. This study aims to detect mango leaf disease by deep learning. Therefore, we proposed a customized EfficientNet model utilizing a channel attention module. In our proposed model vanilla EfficientNet is incorporated with a channel attention module to capture the essential channels of the feature maps to detect mango leaf disease. The experiment is conducted on a dataset consisting of 4000 images of 8 classes. To evaluate the performance, the research is carried out on two segments such as single train-validation-test split and 5-fold cross-validation. Also, the performance of the proposed model is compared with pretrained CNN architectures such as AlexNet and ResNet50. In the result evaluation, it found that the proposed model obtained 98.88% accuracy in a single train-validation-test split and an average 98.93% accuracy in 5-fold cross-validation. It is worth mentioning that the performance of the proposed model is not only superior to AlexNet and ResNet50 but also to methods of related study on the same dataset. Lastly, to provide a model's performance interpretation an explainable AI method (LIME) is used.